Neuronal response during cognitive processing can be mapped in vivo based on vascular and metabolic coupling, assessed using BOLD-fMRI or PET with radiowater, [15O]O2, or [18F]FDG PET. Constant infusion of FDG leads to fairly steady FDG concentration in plasma, after which concentration in tissue increases linearly, revealing the net influx rate of FDG. Monitoring of transient metabolic changes is possible during steady-state FDG PET. These techniques were first applied to study tumour metabolism in small animals (Bérard et al., 2006). When used in human brain activation studies, the steady-state technique was termed as fPET-FDG (Villien et al., 2014). Simultaneous fPET-FDG and BOLD-fMRI can be used to identify task-relevant brain networks (Hahn et al., 2018 and 2020; Jamadar et al., 2019 and 2021), and it has revealed uncoupling of glucose metabolism from the BOLD signal in the default mode network during tasks (Stiernman et al., 2021).

Villien et al (2014) and Hahn et al (2016) have used constant infusion protocol to measure the task-specific glucose metabolism changes in the brain. Using bolus plus constant infusion protocol increases the SNR in the beginning of the PET scan, helping to reduce the necessary task duration to few minutes (Rischka et al., 2018), providing similar or higher reliability than ASL and BOLD imaging (Rischka et al., 2021). However, task-induced change does not affect much the PET signal during the first 2 min, because the increase in FDG trapping is offset by decrease of free FDG (Stiernman et al., 2021). Simple task-resting state comparisons can be performed even after bolus FDG administration (Ripp et al., 2021).

The fPET data can be analysed using similar software that is used to analyse fMRI data. The Brain Imaging Data Structure (BIDS) with PET-BIDS extension (Norgaard et al., 2022) could be an efficient way of storing and organizing fPET data (Jamadar et al., 2020, 2021, and 2022).


Irreversible three-compartment model of FDG

Analysis model is based on irreversible three-compartment model (two-tissue compartment model). FDG concentration in arterial plasma, CA, is the input function, CF is the FDG concentration in interstitial and intracellular brain tissue, and CM is the concentration of phosphorylated (and trapped) FDG in brain tissue. Rate constant K1 represent the unidirectional transport of FDG from plasma compartment across blood-brain barrier into the brain tissue (delivery rate), and k2 represents the unidirectional transport back from tissue to the blood. Rate constant k3 is the rate of FDG phosphorylation in the brain tissue. The radioactivity concentration in brain tissue, CT, is the sum of CF and CM.

The ordinary differential equations for the compartmental model are:

With substitution of CF(t) from equation (1) into equation (2) gives

and substitution of that in equation (3) leads to

, where K1k3/(k2+k3) is the net influx rate (Ki) of FDG.

Constant rate infusion

When FDG is infused at a constant rate, we can assume that after an initial equilibration period the FDG concentration in plasma and the free FDG concentration in tissue are nearly constant during rest, that is, dCF(t)/dt = 0 (Villien et al., 2014). Therefore, during this equilibrium,

and Ki can be calculated from the slope of the tissue TAC (dCT(t)/dt). Further, lumped constant (LC) and glucose concentration in plasma (CPGlu) could be used to convert the Ki of FDG to cerebral metabolic rate of glucose (Villien et al., 2014):

Vascular blood

The equations above ignore the vascular volume inside the brain region of interest (ROI). ROI (or volume of interest, VOI, in 3D) can be drawn on the PET image, and it consists of at least one image pixel (or voxel in 3D). The regional brain TAC is a volume weighted sum of blood and tissue TACs:

, where VB is the vascular volume fraction inside the ROI, and CB is the FDG concentration in blood in that volume. During steady state the FDG concentration in tissue vasculature does not change (dCB(t)/dt = 0). Additionally, VB in the brain is relative low. Therefore, we can approximate that the slopes of regional and tissue TACs are similar during equilibrium.


In a brain activation study including rest and task sessions, during initiation and termination of task the equilibrium is disrupted, but if we assume that a new steady state will be reached, then again dCF(t)/dt = 0, and CMRglu during the new equilibrium can be approximated from the slope of the TAC. The CMRglu could be computed pixel-by-pixel, producing CMRglu maps in the rest and task conditions (Villien et al., 2014).

If the assumption of steady state holds, then the changes in the slope of the tissue TAC are proportional to changes in the rate of FDG Ki and metabolic rate of glucose (Villien et al., 2014). CMRglu change between rest and task conditions,

can be approximated from the change in TAC slopes (dCROI(t)/dt):

If we can assume that LC, plasma glucose concentration, and FDG concentration in plasma are the same during rest and task conditions, those are cancelled out, simplifying the calculation considerably:

While constant infusion of FDG leads to reasonably steady arterial FDG concentration (CA), allowing us to assume that it does not change during the relatively short rest or task stage, CA usually shows upward or downward trend during the PET scan. Therefore it is essential that CA is assessed throughout the PET scan, and, when necessary, the measured FDG plasma concentrations during rest and task are used in the ΔCMRglu calculation. Since only relative quantification of CA is needed, venous blood sampling or image-based blood TAC can be used instead of arterial blood sampling. Use of venous sampling instead of arterial sampling in fPET has been found to provide similar results (Hahn et al., 2016). Sparsely collected blood sampling may require fitting to obtain CA(t) with better quality; in a constant infusion FDG study Hahn et al (2016) fitted two-exponential function to plasma data:

but this is not applicable in bolus plus infusion protocol. Representative plasma and brain grey matter curves after FDG bolus, infusion, and bolus plus infusion are shown by Jamadar et al (2022).

The assumption of stable LC may not be valid, because during brain activation the LC is reported to increase rapidly (Blomqvist et al., 1994); however, accurate LC values are not available, and therefore must be assumed similar, adding unknown bias into results.

If the PET image quality is sufficient, then maps of CMRglu change between rest and task conditions could be computed based on only the slopes of pixel TACs.

Ki maps can be estimated from image data using general linear model (GLM), adopted from fMRI analyses (Villien et al., 2014: Hahn et al., 2016). A baseline regressor has been defined based all grey matter voxels modelled by 3rd order polynomial, with task effects removed by modelling those as nuisance variables. Each task has its own regressor, defined as a linear ramp function with slope of 1 during task and 0 otherwise (Villien et al., 2014: Hahn et al., 2016; Rischka et al., 2018).

Independent component analysis (ICA) is a data-driven method, unlike fixed-model based GLM (Jamadar et al., 2019). In constant infusion FDG study, the FDG concentration in the brain is increasing, possibly also leading also to increased signal-to-noise ratio. The baseline FDG uptake must be removed from the data before ICA unmixing step (Li et al., 2020 and 2021). Inter-subject variance in non-task related FDG PET changes can be removed by normalizing grey matter voxels with individual average grey matter FDG concentration, fitted with 3rd order polynomial (Jamadar et al., 2019). Spatiotemporal gradient filter can be used (Jamadar et al., 2020, 2021), or GLM with a linear regressor that models the continuous baseline increase and the difference between baseline for each image voxel and the grey matter mean (Jamadar et al., 2022).

See also:


Chiaravalloti A, Micarelli A, Ricci M, Pagani M, Ciccariello G, Bruno E, Alessandrini M, Schillaci O. Evaluation of task-related brain activity: is there a role for 18F FDG-PET imaging? Biomed Res Int. 2019:4762404. doi: 10.1155/2019/4762404.

Hahn A, Gryglewski G, Nics L, Hienert M, Rischka L, Vraka C, Sigurdardottir H, Vanicek T, James GM, Seiger R, Kautzky A, Silberbauer L, Wadsak W, Mitterhauser M, Hacker M, Kasper S, Lanzenberger R. Quantification of task-specific glucose metabolism with constant infusion of 18F-FDG. J Nucl Med., 2016; 57(12): 1933-1940. doi: 10.2967/jnumed.116.176156. Erratum in J Nucl Med., 2018; 59(3): 468.

Huang SC, Phelps ME, Hoffman EJ, Kuhl DE. Error sensitivity of fluorodeoxyglucose method for measurement of cerebral metabolic rate of glucose. J Cereb Blood Flow Metab. 1981; 1(4): 391-401. doi: 10.1038/jcbfm.1981.43.

Verger A, Guedj E. The renaissance of functional 18F-FDG PET brain activation imaging. Eur J Nucl Med Mol Imaging 2018; 45: 2338-2341. doi: 10.1007/s00259-018-4165-2.

Villien M, Wey H-Y, Mandeville JB, Catana C, Polimeni JR, Sander CY, Zürcher NR, Fowler JS, Rosen BR, Hooker JM. Dynamic functional imaging of brain glucose utilization using fPET-FDG. NeuroImage 2014; 100: 192-199. doi: 10.1016/j.neuroimage.2014.06.025.

Tags: , , , ,

Updated at: 2023-02-01
Created at: 2022-10-10
Written by: Vesa Oikonen